0. Spacecraft database import (https://www.ucsusa.org/resources/satellite-database)

Spacecraft database is in excel format. Mainly empty columns and columns that contain insignificant data (for the current case) have been removed. Database contains information about spacecraft orbital parameteres, mass, country of origin, launch site, purpose, etc.

import pandas as pd
import plotly.express as px
import matplotlib.pyplot as plt
from sklearn.cluster import KMeans
from sklearn import preprocessing
from collections import Counter

%matplotlib inline

df = pd.read_excel('UCS-Satellite-Database-8-1-2020.xls')

df.head()

	Name	Country	Users	Purpose	Class of Orbit	Longitude of GEO	Perigee	Apogee	Eccentricity	Inclination	Period	Launch Mass	Launch Site	Launch Vehicle	Year of Launch
0	1HOPSAT	USA	Commercial	Earth Observation	LEO	0.0	566	576	0.000720	36.90	96.08	22.0	Satish Dhawan Space Centre	PSLV	2019
1	3Cat-1	Spain	Civil	Technology Development	LEO	0.0	476	500	0.001750	97.40	95.00	4.0	Satish Dhawan Space Centre	PSLV	2018
2	Aalto-1	Finland	Civil	Technology Development	LEO	0.0	497	517	0.001454	97.45	94.70	4.5	Satish Dhawan Space Centre	PSLV	2017
3	AAUSat-4	Denmark	Civil	Earth Observation	LEO	0.0	442	687	0.017665	98.20	95.90	1.0	Guiana Space Center	Soyuz 2.1a	2016
4	ABS-2	Multinational	Commercial	Communications	GEO	75.0	35778	35793	0.000178	0.08	1436.03	6330.0	Guiana Space Center	Ariane 5 ECA	2014

1. Some interesting data

1.1 Top spacecraft operating countries

sc_countries = df['Country'].value_counts()
sc_countries.head(20)

USA               1406
China              375
Russia             170
United Kingdom     129
Japan               80
Multinational       64
India               58
ESA                 53
Canada              39
Germany             33
Luxembourg          32
Spain               21
South Korea         17
Argentina           16
Israel              16
Australia           13
Saudi Arabia        13
France              12
Netherlands         12
Italy               12
Name: Country, dtype: int64

sc_countries.head(20).plot.bar(color='#483D8B', figsize=(15,5), xlabel='Country', 
                               ylabel='Number of spacecraft', title='Top spacecraft operating countries')

<AxesSubplot:title={'center':'Top spacecraft operating countries'}, xlabel='Country', ylabel='Number of spacecraft'>

1.2 Satellite prime purpose

sc_purpose = df['Purpose'].value_counts()
sc_purpose.head(10)

Communications                           1364
Earth Observation                         785
Technology Development                    323
Navigation/Global Positioning             138
Space Science                              88
Earth Science                              13
Navigation/Regional Positioning            12
Technology Demonstration                    9
Communications/Technology Development       8
Space Observation                           8
Name: Purpose, dtype: int64

sc_purpose.head(6).plot.pie(figsize=(10, 10), autopct='%1.1f%%', title='Satellite prime purpose', 
                            colors=['#DDA5B6', '#F2CC8C', '#F1E6C1', '#3F6A8A', '#4D5E72', '#8FB9A8']);

1.3 Launched by year

sc_launchyear = df['Year of Launch'].value_counts()
sc_launchyear.plot(style='d', figsize=(10, 8), title='Launched by year', xlabel='Year', ylabel='Spacecraft launched')

<AxesSubplot:title={'center':'Launched by year'}, xlabel='Year', ylabel='Spacecraft launched'>

1.4 Spacecraft launching sites

sc_launchsite = df['Launch Site'].value_counts()
sc_launchsite.head(15)

Cape Canaveral                     778
Baikonur Cosmodrome                337
Guiana Space Center                276
Satish Dhawan Space Centre         273
Vandenberg AFB                     242
Jiuquan Satellite Launch Center    154
Xichang Satellite Launch Center    124
Taiyuan Launch Center              117
Plesetsk Cosmodrome                114
Vostochny Cosmodrome                64
Dombarovsky Air Base                45
Rocket Lab Launch Complex 1         45
International Space Station         41
Tanegashima Space Center            41
Wallops Island Flight Facility      33
Name: Launch Site, dtype: int64

sc_launchsite.head(15).plot.bar(color='#8FB9A8', figsize=(15,5), xlabel='Launch Site', 
                                ylabel='Number of launches', title='Spacecraft launching sites')

<AxesSubplot:title={'center':'Spacecraft launching sites'}, xlabel='Launch Site', ylabel='Number of launches'>

1.5 Spacecraft users

sc_users = df['Users'].value_counts()
sc_users.head(10)

Commercial               1514
Government                452
Military                  357
Civil                     140
Government/Commercial     114
Military/Commercial        77
Military/Government        57
Government/Civil           43
Commercial/Civil           11
Military/Civil              7
Name: Users, dtype: int64

sc_users.head(4).plot.pie(figsize=(10, 10), autopct='%1.1f%%', title='Satellite main users', 
                          colors=['#DDA5B6', '#F2CC8C', '#F1E6C1', '#3F6A8A']);

2. K-means clastering by orbit class

2.1 Perigee and eccentricity

Let's try to cluster spacecrafts by orbit class using perigee and eccentricity (yes, we already know orbit class thanks to 'Class of orbit' column, but let's do it anyway, just to see if it is possible).

df['Class of Orbit'].value_counts()

LEO           2032
GEO            560
MEO            137
Elliptical      58
Name: Class of Orbit, dtype: int64

Well, there are four types of orbits in our dataset, so there shoud be 4 clasters.

X = df[['Perigee', 'Eccentricity']]

fig = px.scatter(df, x='Perigee', y='Eccentricity')
fig.show()

kmeans = KMeans(n_clusters=4, random_state=13)
kmeans.fit(X)
y_kmeans = kmeans.predict(X)

plt.rcParams['figure.figsize'] = [15, 10]
plt.scatter(df['Perigee'], df['Eccentricity'], c=y_kmeans, s=50, cmap='viridis')
plt.xlabel('Perigee')
plt.ylabel('Eccentricity');

It does not look right at all. Let's try to scale data.

2.2 Perigee and eccentricity, scaled

X_norm = preprocessing.normalize(X, axis=0, norm='max')

kmeans_n = KMeans(n_clusters=4, random_state=13)
kmeans_n.fit(X_norm)
y_kmeans_n = kmeans_n.predict(X_norm)

plt.rcParams['figure.figsize'] = [15, 10]
plt.scatter(df['Perigee'], df['Eccentricity'], c=y_kmeans_n, s=50, cmap='viridis')
plt.xlabel('Perigee')
plt.ylabel('Eccentricity');

So much better! Let's compare our results with dataset.

Counter(y_kmeans_n)  # K-means results

Counter({0: 2063, 1: 562, 3: 118, 2: 44})

df['Class of Orbit'].value_counts()

LEO           2032
GEO            560
MEO            137
Elliptical      58
Name: Class of Orbit, dtype: int64

Not perfect but pretty close.

df_labels = pd.factorize(df['Class of Orbit'])[0]  # set classes of orbits to labels

plt.rcParams['figure.figsize'] = [15, 10]
plt.scatter(df['Perigee'], df['Eccentricity'], c=df_labels, s=50, cmap='viridis')
plt.xlabel('Perigee')
plt.ylabel('Eccentricity');

I'm going to add some features and see if it will improve clustering.

2.3 Perigee, apogee, period and eccentricity, scaled

X_aug = df[['Perigee', 'Eccentricity', 'Apogee', 'Period']]
X_aug_norm = preprocessing.normalize(X_aug, axis=0, norm='max')

kmeans_n_aug = KMeans(n_clusters=4, random_state=13)
kmeans_n_aug.fit(X_aug_norm)
y_kmeans_n_aug = kmeans_n_aug.predict(X_aug_norm)

plt.rcParams['figure.figsize'] = [15, 10]
plt.scatter(df['Perigee'], df['Eccentricity'], c=y_kmeans_n_aug, s=50, cmap='viridis')
plt.xlabel('Perigee')
plt.ylabel('Eccentricity');

Counter(y_kmeans_n_aug)  # K-means results

Counter({0: 2063, 1: 562, 3: 118, 2: 44})

I guess no, the result is the same.